What do we need to have in place before we start?

An MCP server or an OpenAPI spec: one declared surface your agents can reach. If you don't have one yet, it is a smaller job than it sounds and it gets scoped with you. The knowledge-shaped half of what your users ask runs on the context layer and needs no integration at all.

Where does it run, and who holds the data?

It is hosted in your cloud, inside your own service boundary, and self-hosted from the start. Your users' data stays in your systems, which keeps the security review short.

Do our users need a second login?

No. Your app owns login and Studio trusts the identity you hand it, so the person who signed in to your product is the person the agent acts for.

Can the agent write to our systems?

Yes. It reads freely and asks for confirmation before it writes. Per-tenant credential vaulting, tenant isolation and approval gates cover anything consequential, and building that layer properly is most of the work in taking an agent to production.

Who keeps the agents up to date?

The people who own the knowledge. Policies, procedures, pricing and playbooks live in a workspace your team edits directly, and a wording change takes effect as soon as they save it.

How do we know whether it's working?

Every turn is traced: prompt, tools, results, latency and cost, tagged with tenant, user and agent. You can see where the agent guesses, where it stalls and where users give up, which gives you a ranked list of what to fix next.

What happens when we want it inside our main app?

It moves, and nothing is rebuilt. The agents, the tools and the login come with it. The only thing that changes is where the interface renders.

What does it cost, and how long is “weeks”?

Both depend on what you bring. An existing MCP server and a written policy set move fastest; a surface scoped from scratch takes longer. That gets worked out on the call rather than guessed at here.

Intentface Studio — your agentic product, live in weeks

Studio

A real agentic experience in production, running in your own systems under your own brand.

Why did our October bill go up 18 % when we used less electricity than in September?

Called 2 tools

Consumption fell and the bill still rose. What moved is the shape of the day, not the volume.

19 260 kWh in October against 20 070 kWh in September, and the bill went from 2 240 € to 2 640 €. Consumption data
61 % of the month's cost landed in 9 % of the hours — weekday mornings, 06 to 09. Spot prices
The cold spell on 14–17 October pushed those mornings higher at Ruoholahti, where the heating and the kitchen start within the same half hour.

Site	kWh	Cost	Peak-hour share
Ruoholahti	9 140	1 386 €	71 %
Kallio	5 620	702 €	18 %
Pasila	4 500	552 €	11 %

Ruoholahti takes 47 % of the electricity and produces 71 % of the peak-hour cost.

What can we actually do about it?

Called 2 tools

Two moves, largest first.

Shifting hot water at Ruoholahti from 05–07 to 01–04 moves about 640 kWh a month into the cheapest hours — 96 € a month at October prices. Building controls
The kitchen pre-heat starts 25 minutes before the first shift. Ten minutes is enough, and that takes another 180 kWh out of the same window.

Both figures are estimated on hours you actually ran, not on a forecast. A milder October returns less.

Would a fixed contract have been cheaper?

Called 1 tool

No. October would have been 2 710 € on the fixed offer against 2 640 € on spot. Contract

Over the last twelve months the two land within 3 % of each other, and every month spot lost, it lost in the same three morning hours. Moving those hours is worth more than changing the contract.

Ask about a site, an hour or a contract...

Energy desk

What should I be doing in my forest over the next three years?

Called 3 tools

Three stands are approaching thinning, one tending job is time-critical, and one stand is better left alone.

Stand	Area	Class	Action	When
12	4.1 ha	02	First thinning	2027
14	3.6 ha	02	First thinning	2027
21	5.8 ha	03	Thinning	2028
3	1.9 ha	T2	Seedling tending	summer 2027
7	6.2 ha	04	Leave standing	to 2029

Stands 12 and 14 carry 2 140 stems a hectare and share a boundary, so one harvest covers 7.7 ha and the machine move gets paid for once instead of twice. Stand 21 is at 26 m²/ha and can wait the extra year. Stand data

How much timber would come off 12 and 14?

Called 1 tool

About 645 m³ across the two.

Pine pulpwood 310 m³
Spruce pulpwood 240 m³
Sawlog 95 m³

That comes off the 2024 plan's stand data, not a field visit. A bid request produces the figure you can hold someone to, and first thinnings usually come in under the plan rather than over it. Forest plan

Can the seedling tending wait until then?

Called 1 tool

No. Stand 3 is the one job on the list that cannot move.

The birch is about 1.5 m above the spruce and still closing. Tended before summer 2027 the stand recovers; left to 2029 the spruce loses height growth it does not get back. Everything else here has a year of slack in it. Work history

Ask about a stand, a thinning or a bid...

Forest plan

Which machines are likely to cause problems next week, and should I move anything between sites?

Called 3 tools

Two of six. Harvester 04 on trend, Forwarder 02 on interval.

Machine	Site	Hours	Next service	Status
Harvester 04	Kuru	11 240	in 40 h	Attention
Forwarder 02	Kuru	8 905	in 15 h	Due
Harvester 01	Vesanto	14 620	in 310 h	OK
Harvester 06	Vesanto	3 180	in 480 h	OK
Forwarder 05	Vesanto	9 740	in 220 h	OK
Forwarder 08	Kuru	6 310	in 390 h	OK

Harvester 04 runs its hydraulic oil 9 degrees above its own six-week baseline on the same work type, and the drift starts the day after the 21 October hose change. Machine data

Forwarder 02 hits its interval mid-week on a site that finishes Friday, so the cheap window is Thursday, not the Monday after. Service history

Can we keep cutting until Friday?

Called 2 tools

Yes, at this load. The temperature is inside the operating range and nothing has tripped.

Two drifts with the same shape in your own fleet both turned out to be a sticking return valve. Mean time from first drift to fault was 23 days — you are on day 9.

That is a pattern match on your history, not a diagnosis from the machine. It has not reported a fault and I am not claiming one.

What about the site plan?

Called 2 tools

Vesanto is 4 days ahead of plan, Kuru is 2 days behind.

Moving Forwarder 05 to Kuru on Thursday covers Forwarder 02 through its service and pulls Kuru back to plan. It costs nothing extra to move it: the low loader is already booked that day for the Harvester 04 visit. Site plan

Ask about a machine, a site or a service window...

Fleet desk

Rig 14 keeps throwing the same fault. What is it, can we keep operating, and what do we need?

Called 4 tools

Percussion pressure drops below threshold on high feed — 14 times in 6 days, every one of them in the last third of the hole. Fault log

Day	Events	Depth in hole
10 Nov	1	4.1 m
11 Nov	2	3.8–4.2 m
12 Nov	2	4.0–4.3 m
13 Nov	3	3.6–4.2 m
14 Nov	3	3.5–4.1 m
15 Nov	3	3.4–4.2 m

Pressure loss under feed with nothing at idle is accumulator pre-charge loss, not a pump fault. Manual 6.4

Rig 9 threw the same cluster in March. It ran 11 more days and then failed the damper piston. Maintenance history

Can we keep running?

Called 1 tool

Yes, at reduced feed pressure.

At that setting you lose roughly 8 % penetration rate, which is about half a shift across the week. Running unchanged risks the damper piston, and that turns a 2-hour job into a 2-day one.

This is read off the fault pattern and one comparable history. Nobody has looked at the rig, and a pre-charge measurement would settle it in ten minutes.

What parts do we need?

Called 2 tools

Two lines, both covered by your service agreement, so the seal set ships without a separate quotation.

Part	Description	Qty	Availability
87 41 2205	Accumulator charging kit	1	In stock, local warehouse
87 41 2260	Damper seal set	1	5-day lead, order by Thursday

Ordered today, goods receipt at Vuono Mine on Monday 16 November. Parts catalogue

Ask about a unit, a fault or a part...

Service desk

Build

Focus on building your solution.

The UI

Streaming, tool calls, approval prompts, retries and file handling are all built. Send a design guide and the interface wears your brand within hours.

The agent backend

Conversation history, durability, scheduling, approvals and per-tenant isolation, already running in production and maintained for you.

The context layer

Policies, procedures, pricing and playbooks live in a workspace your own team edits. A wording change takes effect as soon as they save it.

Iterate

Start with a small group of real users.

Manage users and access

Invite a team, a segment or one pilot customer, and expand the group when you feel confident. Access is yours to grant and revoke at any point.

Collect production traces

Every turn is tagged with tenant, user and agent, so you can see where the agent guesses and where users give up. The result is a ranked list of what to fix.

Improve it from what you learn

Turn the failures you find into context edits and tool fixes. Each round raises the quality of the next, and everything you learn stays in your own system.

Integrate

It runs inside your own infrastructure.

Hosted in your cloud

Self-hosted from the start, inside your own service boundary. Your users' data stays in your systems, which keeps the security review short.

Signs in through your app

Your app owns login. The user's token becomes the agent's identity, so it acts as that user with exactly the permissions they already have.

Controlled writes to your systems

The agent reads freely and asks for confirmation before it writes anything back. Credential vaulting, tenant isolation and approval gates are part of the platform.

Connect

Connect it to the systems you already run.

Authorised once per organisation

A connection is a URL and a grant, authorised once for the whole organisation. Nobody on your team has to set up an account of their own.

Missing a connector

That happens often, and it is a smaller job than it sounds. It gets scoped and built as part of the project.

Questions need no integration

What's the policy, how does this work, what applies to my case: those run on the context layer, so they ship before any system is connected.

Questions

What do we need to have in place before we start?: An MCP server or an OpenAPI spec: one declared surface your agents can reach. If you don't have one yet, it is a smaller job than it sounds and it gets scoped with you. The knowledge-shaped half of what your users ask runs on the context layer and needs no integration at all.
Where does it run, and who holds the data?: It is hosted in your cloud, inside your own service boundary, and self-hosted from the start. Your users' data stays in your systems, which keeps the security review short.
Do our users need a second login?: No. Your app owns login and Studio trusts the identity you hand it, so the person who signed in to your product is the person the agent acts for.
Can the agent write to our systems?: Yes. It reads freely and asks for confirmation before it writes. Per-tenant credential vaulting, tenant isolation and approval gates cover anything consequential, and building that layer properly is most of the work in taking an agent to production.
Who keeps the agents up to date?: The people who own the knowledge. Policies, procedures, pricing and playbooks live in a workspace your team edits directly, and a wording change takes effect as soon as they save it.
How do we know whether it's working?: Every turn is traced: prompt, tools, results, latency and cost, tagged with tenant, user and agent. You can see where the agent guesses, where it stalls and where users give up, which gives you a ranked list of what to fix next.
What happens when we want it inside our main app?: It moves, and nothing is rebuilt. The agents, the tools and the login come with it. The only thing that changes is where the interface renders.
What does it cost, and how long is “weeks”?: Both depend on what you bring. An existing MCP server and a written policy set move fastest; a surface scoped from scratch takes longer. That gets worked out on the call rather than guessed at here.

Embedded when you're ready.

Book a demo See our work

One 30-minute call to see whether it fits.

Focus on building your solution.

The UI

The agent backend

The context layer

Agents

Start with a small group of real users.

Manage users and access

Collect production traces

Improve it from what you learn

It runs inside your own infrastructure.

Hosted in your cloud

Signs in through your app

Controlled writes to your systems

Connect it to the systems you already run.

Authorised once per organisation

Missing a connector

Questions need no integration

MCP connections

Embedded when you're ready.

Your agentic product, live in weeks.

The UI

The agent backend

The context layer

Manage users and access

Collect production traces

Improve it from what you learn

Hosted in your cloud

Signs in through your app

Controlled writes to your systems

Authorised once per organisation

Missing a connector

Questions need no integration

Embedded when you're ready.